From adapter to deployment: merging LoRA weights and serving with vLLM or a Space

What to do after training: keep or merge the adapter, choose between local inference, a vLLM server and a hosted Space, and check that the served mode